Module ch.randelshofer.fastdoubleparser

Package ch.randelshofer.fastdoubleparser

Class FastFloatMath

java.lang.Object

ch.randelshofer.fastdoubleparser.FastFloatMath

final class FastFloatMath
extends Object

This class complements FastDoubleMath with methods for converting FloatingPointLiteral productions to floats.

See JavaDoubleParser for a description of FloatingPointLiteral.

Field Summary

Fields

Modifier and Type	Field	Description
private static final int	FLOAT_EXPONENT_BIAS	Bias used in the exponent of a float.
private static final int	FLOAT_MAX_EXPONENT_POWER_OF_TEN
private static final int	FLOAT_MAX_EXPONENT_POWER_OF_TWO
private static final int	FLOAT_MIN_EXPONENT_POWER_OF_TEN
private static final int	FLOAT_MIN_EXPONENT_POWER_OF_TWO
private static final float[]	FLOAT_POWER_OF_TEN	Precomputed powers of ten from 10^0 to 10^10.
private static final int	FLOAT_SIGNIFICAND_WIDTH	The number of bits in the significand, including the implicit bit.

Constructor Summary

Constructors

Modifier	Constructor	Description
private	FastFloatMath()	Don't let anyone instantiate this class.

Method Summary

Modifier and Type	Method	Description
(package private) static float	fastScalb(float number, int scaleFactor)	This is a faster alternative to Math.scalb(float, int).
(package private) static float	tryDecFloatToFloatTruncated(boolean isNegative, long significand, int exponent, boolean isSignificandTruncated, int exponentOfTruncatedSignificand)
(package private) static float	tryDecToFloatWithFastAlgorithm(boolean isNegative, long significand, int power)	Attempts to compute digits * 10^(power) exactly; and if "negative" is true, negate the result.
(package private) static float	tryHexFloatToFloatTruncated(boolean isNegative, long significand, int exponent, boolean isSignificandTruncated, int exponentOfTruncatedSignificand)	Tries to compute significand * 2^exponent exactly using a fast algorithm; and if isNegative is true, negate the result; the significand can be truncated.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Details

FLOAT_EXPONENT_BIAS

private static final int FLOAT_EXPONENT_BIAS

Bias used in the exponent of a float.

See Also:

• Constant Field Values

FLOAT_SIGNIFICAND_WIDTH

private static final int FLOAT_SIGNIFICAND_WIDTH

The number of bits in the significand, including the implicit bit.

See Also:

• Constant Field Values

FLOAT_MIN_EXPONENT_POWER_OF_TEN

private static final int FLOAT_MIN_EXPONENT_POWER_OF_TEN

See Also:

• Constant Field Values

FLOAT_MAX_EXPONENT_POWER_OF_TEN

private static final int FLOAT_MAX_EXPONENT_POWER_OF_TEN

See Also:

• Constant Field Values

FLOAT_MIN_EXPONENT_POWER_OF_TWO

private static final int FLOAT_MIN_EXPONENT_POWER_OF_TWO

See Also:

• Constant Field Values

FLOAT_MAX_EXPONENT_POWER_OF_TWO

private static final int FLOAT_MAX_EXPONENT_POWER_OF_TWO

See Also:

• Constant Field Values

FLOAT_POWER_OF_TEN

private static final float[] FLOAT_POWER_OF_TEN

Precomputed powers of ten from 10^0 to 10^10. These can be represented exactly using the float type.

Constructor Details

FastFloatMath

private FastFloatMath()

Don't let anyone instantiate this class.

Method Details

tryDecFloatToFloatTruncated

static float tryDecFloatToFloatTruncated(boolean isNegative,
 long significand,
 int exponent,
 boolean isSignificandTruncated,
 int exponentOfTruncatedSignificand)

tryHexFloatToFloatTruncated

static float tryHexFloatToFloatTruncated(boolean isNegative,
 long significand,
 int exponent,
 boolean isSignificandTruncated,
 int exponentOfTruncatedSignificand)

Tries to compute significand * 2^exponent exactly using a fast algorithm; and if isNegative is true, negate the result; the significand can be truncated.

Parameters:

isNegative - true if the sign is negative

significand - the significand (unsigned long, uint64)

exponent - the exponent number (the power)

isSignificandTruncated - true if significand has been truncated

exponentOfTruncatedSignificand - the exponent number of the truncated significand

Returns:

the double value, or Double.NaN if the fast path failed.

tryDecToFloatWithFastAlgorithm

static float tryDecToFloatWithFastAlgorithm(boolean isNegative,
 long significand,
 int power)

Attempts to compute digits * 10^(power) exactly; and if "negative" is true, negate the result.

This function will only work in some cases, when it does not work it returns null. This should work *most of the time* (like 99% of the time). We assume that power is in the [-325, 308] interval: the caller is responsible for this check.

References:

Noble Mushtak, Daniel Lemire. (2023) Fast Number Parsing Without Fallback.

arxiv.org

Parameters:

isNegative - whether the number is negative

significand - uint64 the significand

power - int32 the exponent of the number

Returns:

the computed double on success, Double.NaN on failure

fastScalb

static float fastScalb(float number,
 int scaleFactor)

This is a faster alternative to Math.scalb(float, int).

This method only works if scaleFactor is within the range of Float.MIN_EXPONENT through Float.MAX_EXPONENT (inclusive), so that we do not underflow or overflow.

Parameters:

number - a double number

scaleFactor - the scale factor

Returns:

number × 2^scaleFactor